xiejunjie524/handdraw-story-video

handdraw-good-deed-story

Batch-produce 35–45 second vertical hand-drawn good-deed story videos with black line art revealed left to right and aligned low-saturation color filling in the same direction.

Zobacz źródło

Jak używać

  1. Skopiuj polecenie instalacji i sprawdź źródła przed instalacją we własnym środowisku agenta.
  2. W oryginale sprawdź wymagane narzędzia, dane dostępowe i obsługiwane środowiska.
  3. Opisz dane wejściowe i oczekiwany rezultat; porównaj wynik ze źródłami.
npx skills add https://github.com/xiejunjie524/handdraw-story-video --skill handdraw-good-deed-story

Przed rozpoczęciem

  • Instalacja nie obejmuje subskrypcji modelu, limitu API ani uprawnień do plików.
  • Gwiazdki i instalacje pokazują aktywność, a nie niezależny test jakości lub bezpieczeństwa.
Oryginalny dokument Skill

Treść z repozytorium z zachowaniem nagłówków, przykładów, kodu, tabel, linków i obrazów.

Hand-drawn good-deed story

Create a compact cause-and-effect story, then build it with the bundled scripts and HyperFrames.

Defaults

  • Deliver 40 seconds, 720×960, 30 fps, H.264 MP4.
  • Use 8 distinct beats of about 5 seconds; accept 7–9 beats and 35–45 seconds.
  • Use no narration by default. Select licensed BGM for each story instead of fixing one track.
  • Generate one 1K low-saturation color mother image per beat. Never retry image generation automatically.
  • Derive line art locally from the exact color image with scripts/make_lineart.py; never generate line and color images separately.
  • Reveal line art left to right, then fill color left to right from the same origin.
  • Render with one worker and never overwrite a prior render.

Workflow

  1. Read docs/story-spec.md and write an eight-row beat sheet: event, characters, setting, key prop, action/reaction, two visual details, and framing.
  2. Keep every beat visually distinct. Reject repeated mother images, crops, mirrors, recolors, or zooms used to pad runtime.
  3. Copy templates/story-template.json to a project story.json and fill it in. Manually wrap captions; do not use HTML line breaks.
  4. Generate one complete color mother image for each beat, sequentially, at 1K. Keep broad paper-white negative space and clean separated contours.
  5. Run python scripts/make_lineart.py COLOR OUTPUT_LINE for every approved mother image.
  6. Run python scripts/build_story.py story.json hyperframes/index.html --check-assets.
  7. Install GSAP through npm and copy node_modules/gsap/dist/gsap.min.js to hyperframes/assets/vendor/gsap.min.js.
  8. Run HyperFrames checks, inspect the line and color midpoint of every scene, then render with --workers 1.
  9. Verify duration, dimensions, audio stream, black frames, text fit, consistent reveal direction, and final color before delivery.

Visual constraints

  • Keep the illustrated action mainly in the lower 45%–55% when captions are present.
  • Use at most 2–3 people and two key background anchors per frame.
  • Avoid dense architecture, crowds, heavy rain hatching, and line-art extraction that becomes a black mass.
  • Vary framing: establish place, show a prop/detail interaction, show action/reaction, and end on a changed communal image.
  • Keep captions optional, at 1–3 lines and no more than about 18 Chinese characters per line.
  • Add in-scene text only when a physical note, sign, phone, label, or receipt exists in the story.

Use the builder rather than rewriting mask and timeline boilerplate.

Powiązane zadania